The Company

Founding

Vast.AI (technically Vast.ai, all lowercase, but rendered Vast.AI for clarity in writing) was founded in 2016. The founding context is essential: 2016 was years before ChatGPT, before the LLM boom, before "AI compute" was a venture category. The original users were academic ML researchers, computer-vision practitioners, and crypto miners who needed GPUs for either training or proof-of-work hashing.

The founding team came from technical backgrounds — Greg Diamos had been at Baidu's Silicon Valley AI Lab and prior to that NVIDIA, working on CUDA. He'd seen first-hand how GPU compute was used inside large research labs and recognized the inefficiency: GPUs in academic labs sat idle most of the time; GPUs at hyperscalers were rented at 10x what they cost.

The company has been notably quiet about itself over the years — minimal PR, founder rarely on conference circuits, no big funding announcements. That posture is unusual in the AI-infrastructure space, where most companies cultivate visibility.

The original thesis

Three claims, all of which have proved out:

1. GPU compute will get more important, not less. Deep learning was just starting in 2016; the team bet it would scale. They were right by orders of magnitude.
2. Hyperscaler GPU pricing reflects margin, not cost. AWS, Azure, and GCP charge GPU rates well above the marginal cost of operating the hardware. That gap is room for a marketplace.
3. Supply already exists — it's just not aggregated. Universities, small datacenter operators, mining farms (back when crypto-mining GPUs were a thing), and individual enthusiasts had GPU hardware sitting idle. A marketplace could turn that into rentable capacity without anyone building new datacenters.

The third point is what makes Vast structurally different from a traditional cloud. Vast doesn't operate GPUs. Vast operates a marketplace. The GPUs belong to providers — third parties who list their hardware on Vast in exchange for a share of the rental revenue.

This is a profound architectural difference from CoreWeave, Crusoe, Lambda, or any traditional cloud. Those companies own the hardware. Vast does not. The implication is that Vast can scale supply without capital expenditure — but Vast also has limited control over the quality of the supply on its platform.

Scale today

Vast has not published precise figures, but the public signals point to a marketplace at meaningful scale:

• Tens of thousands of GPUs listed at any given moment. The actual number fluctuates as providers join and leave; the marketplace dashboard exposes live availability.
• Hundreds of thousands of unique customer accounts over the platform's history, with active monthly users in the tens of thousands.
• Coverage from consumer cards (RTX 3090, 4090) up through datacenter-grade (A100, H100, H200, B200). The mix tilts more consumer than enterprise neoclouds, but the high end is well-represented.
• Revenue at meaningful but undisclosed scale. Public estimates have put Vast at meaningful eight-figure ARR; the company hasn't confirmed.

By absolute GPU count or revenue, Vast is much smaller than CoreWeave or Crusoe. By number of unique users and breadth of available hardware models, Vast is one of the largest GPU markets in the world. The two metrics measure different things; both are real.

Ownership & funding

Vast.AI has been notably under-the-radar on funding. Public records show seed-stage funding; the company has not raised the kind of large growth rounds that CoreWeave, Crusoe, Lambda, or Together.AI have done.

The reasonable inference is that Vast is profitable or close to it on a marketplace-take-rate business model, which doesn't require the kind of capital those competitors raise to buy GPUs. The marketplace is asset-light by design.

This shapes the strategic posture: Vast doesn't need to grow revenue at venture pace; it doesn't have a board pushing for an IPO timeline; it can be slow and deliberate about strategic moves. That's an unusual position in the 2026 neocloud landscape where most competitors are either public (CoreWeave, Nebius) or pushing hard toward exits.

Organizational shape

Small. Vast has historically operated with a headcount in the low tens of engineers, with a flat structure and a strong technical bias. The team's public profile is light; LinkedIn shows a company that's never grown into a sprawling org.

The leanness is intentional. A marketplace at Vast's scale can be operated by a small team — the customers (both supply and demand sides) self-serve through the web UI and API. Vast's job is to keep the marketplace running, improve trust and ranking systems, and resolve disputes. It's not a sales-led business.

Compare with CoreWeave at 1,000+ employees or Lambda at several hundred. Vast does similar gross-merchandise-volume work with a fraction of the headcount because the business model doesn't require sales teams, customer success managers, or datacenter operations staff.

Culture & posture

From the outside, Vast presents as engineering-led, anti-marketing, comfortable with niche positioning. Behaviors that signal this:

• The website is functional rather than designed for marketing impact.
• The product surface (CLI, API, web UI) is technical-first; no consumer onboarding flow.
• Documentation is dense and assumes Linux + SSH + Docker familiarity.
• Public communication is rare — releases land without big announcements.
• The community on Discord and Reddit is active and helpful, but it's not company-managed evangelism; it's organic.

This culture has advantages and disadvantages. It keeps Vast cheap to operate, attracts a self-selecting technical user base, and avoids the trap of over-promising. It also limits Vast's ability to capture enterprise customers, who expect polish, account managers, and SLAs.

Takeaway

Hold these facts as you read the rest of the guide:

• Vast is a marketplace, not an operator of GPUs. The implications cascade through every chapter.
• The company is small, deliberate, and structurally unlike its better-funded competitors.
• The original 2016 thesis has proven out, but the strategic question now is whether marketplaces can scale into the enterprise segment or whether they're permanently capped at indie / research / experimental workloads.

The next chapter pulls apart the marketplace model itself.